py-easy-mcp ⚡🤖

Ultra-lightweight, zero-dependency Python framework for Model Context Protocol (MCP). Build MCP servers for Claude in seconds.

Turn standard Python functions (both sync and async def) into Model Context Protocol tools, resources, and prompt templates for Claude Desktop and Claude Code with simple decorators.

Start here: your first contribution

Featured beginner issue #11: Write an offline walkthrough of the existing math example. Read CONTRIBUTING.md for setup, claiming an issue, and opening a draft PR.

Repository: easy-mcp; PyPI distribution: py-easy-mcp; Python import: easy_mcp. Both py-easy-mcp and easy-mcp are installed CLI aliases.

Try this from a reviewed source checkout, in the repository root, with Python 3.8+. It uses synthetic inputs and needs no API key or network access:

from examples.basic_math import mcp
print(mcp.call_tool("add", {"a": 2, "b": 3}))

Expected output:

{'content': [{'type': 'text', 'text': '5'}], 'isError': False}

Related MCP server: MCP Python Server — API Wrapper

⚡ Why py-easy-mcp?

• Zero dependencies: Written in 100% pure standard Python. No third-party runtime dependencies; review code and build tools before use.

• Sync & Async Support: Supports both standard def and modern async def tool, resource, and prompt handlers.

• Full MCP Protocol Support (v0.3.0): Tools (@mcp.tool()), Resources (@mcp.resource()), and Prompts (@mcp.prompt()).

• 1-Click Claude Desktop Installer: Run py-easy-mcp install server.py to auto-detect and configure claude_desktop_config.json with automatic backup.

• Dev Inspector & Interactive REPL: Test your tools in the terminal with py-easy-mcp dev server.py without needing Claude Desktop open.

• Automatic Schema Inspection: Docstrings and type annotations are automatically converted into standard JSON Schemas.

• Offline & Safe: Standard I/O (stdio) JSON-RPC 2.0 communication.

📦 Installation

pip install py-easy-mcp

(Note: Both CLI commands py-easy-mcp and easy-mcp are available)

🚀 30-Second Quickstart

1. Write your server (my_server.py)

from easy_mcp import EasyMCP

mcp = EasyMCP(name="my-server", version="0.3.0")

# 1. Register a Sync or Async Tool
@mcp.tool()
async def fetch_weather(city: str) -> dict:
    """Fetch weather forecast for a given city.

    Args:
        city: The city name (e.g. 'Tokyo', 'San Francisco')
    """
    return {"city": city, "temp": "22°C", "condition": "Sunny"}

# 2. Register a Resource (readable context)
@mcp.resource("memo://guidelines")
def guidelines() -> str:
    """Engineering principles."""
    return "1. Keep it simple.\n2. Zero external dependencies.\n3. Write clear tests."

# 3. Register a Prompt Template
@mcp.prompt()
def code_review(code: str, language: str = "python") -> str:
    """Review code for bugs and maintainability."""
    return f"Please review this {language} snippet:\n\n{code}"

if __name__ == "__main__":
    mcp.run()

2. 1-Click Install to Claude Desktop

Simply run:

py-easy-mcp install my_server.py

py-easy-mcp will:

1. Locate your OS Claude config (%APPDATA%\Claude\claude_desktop_config.json on Windows, ~/Library/Application Support/Claude/... on macOS).

2. Create a timestamped backup (claude_desktop_config.json.<date>-<time>.bak) before changing an existing config.

3. Register my-server pointing to your current Python environment.

If another script is already registered under the same name, the install stops instead of replacing it. Pick a different name with --name, or pass --force to replace it.

Restart Claude Desktop, and your tools will appear instantly!

🛠️ CLI Utilities & Dev Tools

Logging from tools

stdout carries the JSON-RPC stream, so while the server runs, print() inside a tool is sent to stderr, where Claude Desktop keeps its server logs. Error details and tracebacks also go to stderr; the model only sees the error type and message.

Dev Inspector (Offline Testing)

Inspect all registered schemas, parameter requirements, and docstrings:

py-easy-mcp dev my_server.py

Direct Tool Execution

Test a tool directly from your command line:

py-easy-mcp dev my_server.py --call fetch_weather -a '{"city": "Tokyo"}'

Interactive REPL

Start an interactive terminal session to explore tools and resources:

py-easy-mcp dev my_server.py -i

Scaffold a New Project

py-easy-mcp init weather-server

📂 Production MCP Examples

Check out our ready-to-run examples in examples/:

• examples/sqlite_server.py - SQLite database inspector with list_tables, describe_table, execute_query, and a schema DDL resource. It opens SQLITE_DB_PATH read-only and stops queries that run longer than 5 seconds.

• examples/basic_math.py - Arithmetic calculation tools.

• examples/pii_shield_mcp.py - AI data privacy sanitizer for Claude.

🤝 Contributing for Hacktoberfest

We welcome beginner-friendly contributions! Add new lightweight tool examples or protocol extensions. See CONTRIBUTING.md to get started.

📄 License

MIT License © 2026 lui01212